A simultaneous multi-mode identification and outlier filtering method for imbalanced nonlinear process monitoring

Process monitoring is essential for ensuring the safety, reliability, and efficiency of industrial production processes. However, traditional process monitoring methods struggle with multi-mode processes simultaneously containing outliers, especially when the data are nonlinear and imbalanced. To address these challenges, this paper proposes a novel nonlinear process monitoring method that combines improved Connectivity Kernel based Density Peak Clustering with Outlier Filter (CKDPOF) technique and Cost-sensitive Support Vector Data Description (CSVDD). The core contributions of this study are twofold. First, we develop a CKDPOF method that integrates a connectivity kernel technique for identifying data manifolds with a local center extraction strategy aimed at clustering modes and filtering outliers. Second, we propose a CSVDD model that enhances SVDD by incorporating semi-supervised learning concepts, effectively leveraging available anomaly information to create a highly discriminative model capable of mitigating the negative impact caused by imbalanced data. It is particularly noteworthy that the collaborative relationship between CKDPOF and CSVDD can enhance the robustness of fault detection and improve the accuracy of modal identification. Extensive experimental conducted on a simulated wastewater treatment plant platform conclusively demonstrate the superiority of the proposed method in terms of various evaluation indices.